Question on NVIDIA GPU Inference Engine (GIE)


It seems that GIE only output the final results of a caffe trained network.
How can I access the intermediate conv layers output?

For example, in face recognition CNN is used as feature extraction. Multiple intermediate conv layers output are used as extracted feature, which are further processed by PCA, LDA or other metric learning methods. In another example, object detection like faster-RCNN or SSD, rois are estimated from conv layers output.

Is there a plan to make these intermediate conv layers accessible in the final release of GIE?

Don’t use GIE, just use Caffe or whatever framework you are normally using for the inference phases.